Multi-pack battery voltage monitor

Oops! ^^^^^--> not!

Grant.

Wow! What a turnout!!

I apologize for my apparent disappearance. I had to change new readers before I could reply back. This also caused me to lose all posts between 8:00 am and 3:00 pm on July 22; so you wrote something in that window, I fear I can't see it.

I don't have the answer to every question, but I'll give it my best shot.

1. I used the word 'pack' to refer to a 6-cell, lead-acid battery--the same type as you use to start your car. Open-circuit voltage about 12.6 V.

If there's an unambiguous industry term for that, I wouldn't mind hearing it.

A string of 72 would add up to 907 V. I have worked with busses close to 700 V, but this may be a record, especially if it triggers some kind of UL rule.

1. By 'offset' I presume you (PeterD) mean some kind of voltage regulator? Cost is a factor and I should explain the scenario.

The appliance is a UPS. The batteries must be tested at regular intervals. The UL spec calls for measuring the voltage of the entire stack and warning the user when the average voltage per cell drops below a certain threshold. But real-world batteries don't deteriorate uniformly and we would like to measure them individually and sound an alarm when a single battery can not pull its own weight.

This happens while the unit is driving its full rated load on the battery, so the voltage will be dropping as we go. The shortest test lasts 30 seconds, and fitting 72 tests into 10% of this window gives us 42 msec/conversion (assuming they are executed sequentially). If that proves a burden, I can also make the test longer. I don't want it to take so long that the last battery looks bad only because it was measured way down the discharge curve.

1. I don't have a spec for accuracy, but let's say we sound an alarm when a battery drops to 10 V. If I put 10 bit steps between 12 & 10; that's 0.2 V. That's crude but a starting point. I just realized I will have to account for the accuracy of my voltage regulator if it's in series with the A/D. Getting complicated.

I would have to look at the curves for zener diodes. That's obviously cheaper than a regulator, but I lose more accuracy.

1. My boss wants to use a bank of resistive dividers enabled one at a time by relays into a single A/D, referenced to ground (similar to a recommendation posted here). I don't like this as the top battery will be subject to a large common-mode voltage, making that measurement much less accurate than the bottom battery's.

1. I don't have to control the current or voltage of the batteries, just want to monitor their health.

2. Whatever solution I implement will be under the control of a microprocessor. Must execute without human intervention.

1. I believe I can draw milliamps out of the batteries during the test. The challenge is the quiescent current, as many a customer will take delivery and stash it in a warehouse for 6 months before trying to install it, and the test circuit may not load it down while in storage.

2. I don't mind using opto-couplers, but 72 of them will add up fast.

1. PIC does make a uP with built-in A/D converter and UART for under . Isolation is still unresolved. We definitely cannot afford 72 isolated power supplies.

Hope that gives you a better picture. ============================================================ Gary Lynch To send mail, change no$pam gary.lynch@no$pam.com in my domain name to ieee. ============================================================

For low power option, described upthread.

Grant.

Use groups.google.com and then select "Advanced Search" select some keywords and limit search to sci.electronics.design.

Use a separate optocopler to fire up the test.

Use a modular design, with say 9 batteries served by an isolated module. Depending on the physical battery layout, some other grouping might simplify cabling requirements.

Anyway, the number of batteries is so low that dividing down the voltages from these 9 batteries can be done without extremely critical component tolerances. The processor in each module would require a 9 input multiplexor to select one of the voltage dividers at a time to the ADC.

With eight isolated modules, the number of optoisolators is reduced, if the modules are addressed separately, a single series current loop can be used to read the measurements. A simple ON/OFF addressing would require eight separate wire pairs to the module command port. However, if the select signal contains an address, only a single loop would be required to command all modules.

Normally, the command loop would be powered down, which would also put the modules to sleep. A common broadcast command would activate all modules to start measuring. Later on, each module would interrogated and each module would send the stored 9 measurements in a single message.

Thus, only two pairs of wires would be needed from the main processor to the battery array, one for the command loop and the other for reading the results.

A modular design is more flexible, allowing the same modules to be used with different number of batteries.

On a sunny day (Fri, 23 Jul 2010 21:16:40 -0400) it happened "Gary Lynch" wrote in :

One can use a small 1 transistor RF generator, and 72 secondary 2 turn windings or maybe even extremely small ferrite toroids with each a diode and capacitor as rectifier to make say 3.3V for each PICs. You could use optos to send the data back, or perhaps simply have each PIC, as you only need an 'alert', do a PWM output on same ringcore at a different frequency and detect that elsewhere,

Or just have each PIC load the supply by switching in a low value resistor, you can then detect the amplitude variation at the oscillator that feeds them, like this:

impedance matching ________________________________________________________________________________ | | osc - core0 ---wire -- core1--- core2 --- core3 --- core4 - - core72 -- detector core-- | | | | rect1 rect2 rect3 filter ----| ----| | | PIC1 | PIC2 detector | | | | | | R1 | R2 comparator |____| |____| | alarm In the same way you can send the data back if you do not like optos. I mean each PIC has an output that in case of alarm loads the supply with a specific low frequency by switching in a low value resistor (R1, R2, ... ). The current variation is seen in the chain, and that frequency filtered out and detected, and triggers the alarm. You could even have a different pattern for each PIC.

Usenet patent All Rights Reserved. Copyright Jan Panteltje 2010-always, hereby donated into the public domain.

Sorry, don't see it and don't remember it. Got a pointer?

The trip point is 600 V nominal, from 29USC1910 Subpart S.

It may be a bit challenging to operate relays that fast. Relays with that much withstand / isolation are not known for being fast. Look for solutions running multiple converters and mux the data streams as well.

And you get clobbered on temperature stability, micro and isolator per

12 V pack sounds better and better.

See comments for 2 and 3.

OK

OK

Lead-acid batteries are quite notorious for high self discharge.

Not nearly as fast as 1 kV relays.

I would leach off the 12 V packs individually, it will be easy to get the standby current below 0.1 of self discharge. Even active current will be as much for the optos as the uC.

as rectifier

frequency and detect that elsewhere,

like this:

________________________________________________________________________________

|

detector core--

|

filter

|

detector

|

comparator

|

alarm

specific low frequency by

detected, and triggers the alarm.

That is a slick idea. Add a blinking led to each PIC and pwm the led to get the modulating signal. Isolation, general alarm, and specific battery identification all with the same parts.

Ok, with 72 batteries, you'll have 144 connections to make, even if you do not use any measuring/monitoring. So, if you do decide to set it up for measuring/monitoring, it makes sense to integrate the design of the battery to battery connectors with the connections that will be needed for measuring/monitoring.

You want the time taken to connect those 72 batteries to be the same whether or not measuring/monitoring will be included. In effect, the inter-battery connectors will have a "port" on them that allows easy/fast connection of the whatever measuring monitoring equipment. The cost of the measuring equipment could easily be exceeded by the cost of connecting it, if you don't plan up front. Conceptually:

-------- -------- | | | | | --,______________,-- | | --' ^ '-- | | Batt 1 | | | Batt 2 | | | P | | } } } }

If you do that, you could build a resistor into the port so that there is no high current path in the event the wires connected to the port short to ground or other port connected wires.

The actual "port" could be nothing more than an extra wire crimped to the connector you attach to the battery, or a wire with a ring terminal (which would add a bit to assembly time) if the battery uses nut & bolt connections.

There may be many variations on the theme, but bottom line you'll need 74 connections to be able to measure each battery individually and you need to consider how to make them not only effectively, but cheaply.

Ed

=_NextPart_000_0010_01CB2F6B.CD3686E0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable

I have been trying to flesh out some of the circuits discussed here. The following will take us beyond my ability to draw circuits in ASCII, so please refer to the graphic at=20

-

beginning with Circuit A.

I liked the idea of drawing power from the battery, only while the measurement is in progress, and inserted my circuit in series with a 7.5 V zener diode (7 V isn't that common). The A/D will not accept input voltages greater than the power to the chip, so two more resistors are needed to scale it down.

For my voltage-to-frequency converter circuit I favored one microprocessor over a fistful of analog components and MicroChip offers the PIC10F220 with an 8-bit A/D and an on-chip timer, so I can generate a variable frequency or variable duty cycle, or whatever waveform best suits the application. I can buy it from Digi-Key for $0.36 (in onesies).

Finally I can shift the information down to logic levels with an opto-coupler. (An H11A1 costs $0.19.)

But my micro draws up to 80 mA, so I decided to shut down the circuit when not in use, and I decided to deploy the PB1168-ND introduced by Ed. At $3.19, this costs more than the rest of the circuit combined.

But closer inspection of the micro discloses that it uses Vdd as the A/D's reference voltage. Letting that swing through several volts during the test (even though the micro can apparently take it), adds too much complexity, so I have to modify the 'voltage regultor.'

In Circuit B, I have moved the zener down to the circuit common and lowered the voltage to 4.7 V. This might be worth building.

Almost forgot: I mis-read the spec for battery series/parallel combinations. The tallest stack contains 35 packs, so the max voltage is only 440 V. This might allow a cheaper switch than the relay.

Did I miss anything? =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D Gary Lynch To send mail, change no$pam gary.lynch@no$pam.com in my domain name to ieee. =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D

My suggestion of using a 7 V zener was in place of R1. However, a 10 V zener would be better, since it would translate the battery voltage range 10-15 V (charge/discharge) into the ADC input 0-5 V range. Thus a fewer number of bits would suffice. For a larger swing, the R1 would also be needed.

If you use some kind of uP, bit banging some serial data would be as easy and avoid the inaccuracy caused of the PWM detection at the central controller.

Since you have plenty of voltage available, why not use some series regulator power supply, which is remote controlled by an other optoisolator, avoiding the expensive relay.

A couple of points.

If you've decided to use the PB1168-ND relay anyway, as your post indicates, there is no need for all the parts in your diagram. You can just put a resistor in series between the relay contact and the battery for a high impedance connection to your single measuring/ monitoring circuit. Mounted at the battery, that solves the fuse/short circuit in the wiring/short across the battery issues that were mentioned in the thread. There was also a question or relay reliability raised in the thread. That relay is spec'd at > 10^8 mechanical operations, and 5 x 10^5 electrical operations at 125VDC at .24 amps.

Maximum transfer + bounce time is 8 mS. With any relay, you want a diode across the coil. That delays the drop out time on the PB1168-ND relay for a maximum of 5 mS. So given your 42 mS spec, you can program to select the relay, wait 9 mS, measure the voltage for 27 ms, wait for

6 mS then select the next relay. A similar scheme should be used if you use any relay. You should adjust the timing as necessary for the specific relay chosen.

If you go with SPST as shown in your diagrams, you can use reed relays to save money. For example, Digikey 306-1063-ND for $1.43 in qty 10, and it includes the diode across the coil. There are cheaper ones - look around. I think you can get it to less than a dollar per relay by using SMT.

I'd be concerned with the labor cost of assembling and calibrating

35 of those circuits. You'd certainly want a single program replicated in each PIC10F220, which implies that any calibration would be futzing with the ADC input voltage divider. That would be cost prohibitive, so the assumption is that there will be no calibration. If you simply pass the voltage from the battery to a single PWM or measurement device, then you can calibrate it once, but that gets back to needing 2 relays or a double pole relay per battery - or a 2 fet/inverter circuit.

It's easy to get "seduced" by the electronics. Just don't overlook the labor cost of whichever approach you investigate.

Ed

Yo, ($%^#$*%^*) lose the html, this is usenet.

On Fri, 30 Jul 2010 17:50:46 -0400, ehsjr wrote: [snip]

[snip]

Is that last time, 6 ms, intended for transmitting the reading back to a controller?

If multiple relays can be on at the same time (eg, relay drivers controlled by bits from a SIPO shift register, not multiplexed) one can overlap all but the readout parts of the process.

Eg, at time 9*k ms engage relay #k; at 9+9*k, begin measuring voltage at station #k; at time 36+9*k, begin readout #k; at time 42+9*k, open relay #k. For 35 stations, k = 0 to 34 and process is done at time 348 ms.

This approach has a disadvantage as the A/D built into the micro uses Vdd as a voltage reference. As the battery terminal voltage degrades with age, it will cause my A/D readings to shift as well. This will render my data worthless unless I can hold Vref steady over a wide range of battery voltages. That is why I switched to shunt regulation.

Because the divider network draws power from the battery all the time. The relay as drawn removes all load when the circuit is not in use. I am shopping around for a cheaper device. I think I can beat $3.19. ============================================================ Gary Lynch To send mail, change no$pam gary.lynch@no$pam.com in my domain name to ieee. ============================================================

No. That allows time for the relay to drop out.

No. The design allows energizing only 1 relay at a time.

Sorry I wasn't able to reply sooner, I was travelling.

Ed

(eg, relay drivers

Ooops - regarding my previous answer, I had a different circuit in mind. The dropout time comment is valid, but he could select more than one relay at a time with the circuit he posted.

Ed